Distributed DNN based Processing for Uplink Could-RAN

This paper investigates the uplink reception in the cloud radio access network (C-RAN) with finite-capacity fronthaul links. The latter is an emerging network that transfers the computing load from the radio heads (RHs) to the central processor (CP) unit. Due to the prohibitive complexity of computa...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2021 IEEE International Symposium on Information Theory (ISIT) S. 3092 - 3096
Hauptverfasser: Zhang, Chao, Askri, Aymen, Othman, Ghaya Rekaya-Ben
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 12.07.2021
Schlagworte:
Artificial neural networks
Computer architecture
Decoding
Deep learning
Distributed processing
Quantization (signal)
Receivers
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper investigates the uplink reception in the cloud radio access network (C-RAN) with finite-capacity fronthaul links. The latter is an emerging network that transfers the computing load from the radio heads (RHs) to the central processor (CP) unit. Due to the prohibitive complexity of computations, the most efficient uplink C-RAN schemes are challenging to be implemented in practical systems. Using deep neural networks (DNNs), we propose a new and low complex distributed processing for uplink C-RAN subject to some quantification rules. The objective of our architecture, called TDNet, is to optimize the processing jointly at the RHs and the CP side. Our goal is not to solve signal detection in multi-antenna systems. Instead, our work aims to find a helpful transformation scheme at the RH side before quantization. A correspondent decoding scheme at the CP side considers the quantization scheme. Inspired by the projected gradient descent algorithm, TDNet is designed as a distributed DNN with sparse connections. Numerical results are provided and show that our scheme outperforms linear receivers such as the zero-forcing (ZF). It also achieves near-optimal performance compared to the sphere decoder (SD) algorithm, especially for a low-to-moderate number of quantization bits.
DOI:10.1109/ISIT45174.2021.9518195